Understanding How AI Fits into Clinical Pharmacy

Artificial intelligence (AI) may currently have a limited role in pharmacy practice, but that's expected to change as healthcare continues to move toward digital solutions. Now is an ideal time for the pharmacy workforce to catch up on how AI might work in the near future - and where it might fall short, said Andrea Sikora, a clinical associate professor at the University of Georgia College of Pharmacy.

Sikora is helping lead two sessions at ASHP Pharmacy Futures 2024, Infrastructure Needs for Applying Artificial Intelligence to Clinical Pharmacy and Glimpses of AI at the Bedside. She provided a sneak peek of the two sessions for ASHP News Center.

“The key learning I hope attendees will take away is the incredibly important need for infrastructure to utilize artificial intelligence, machine learning, in the medication-use space,” Sikora told ASHP TV.

Demystifying AI

• AI “is a very broad term, which includes machine learning, which includes logistic regression,” said Sikora. “We use regression in basically every study of drug therapy, so clinical pharmacists are using AI whether they realize or not.”
• More advanced machine learning models are becoming a standard for prediction and scientific inquiry. AI for clinical decision support systems, however, is still new, she said “and there is limited validation for drug therapy, so this is a more-to-come area.”

Limited bedside use of AI — for Now

• Though AI use remains in early stages for bedside care, Sikora says, there are a few examples already in use, including AI-based alert systems to improve the rapid identification of septic patients and appropriate interventions.
• Challenges to implementing AI include a lack of common data models allowing electronic health record data to "talk" among institutions; lack of accepted clinical metrics for use of models; and a lack of standardization for ongoing use of the models, said Sikora.

Taking a deeper dive into ICU medications and AI models

• At her session on infrastructure needs, Sikora will discuss how AI models exclude these medications from their datasets. ICU medications, for example, pose unique, nuanced challenges. The average ICU patient, she noted, has 13 medications ordered at any given time.
• Sikora was a co-author of a study that found unsupervised machine learning, in combination with a common data model, detected patterns among the medication regimens of 1,000 ICU patients. Such results have the potential to be used in guiding medication-related treatment decisions.

But first, critical infrastructure needs

• Digital datasets should meet the metrics known as FAIR, which stands for findability, accessibility, interoperability, and reuse-ability. Organizations must develop ethics standards and guidelines, said Sikora.
• Sikora also cited a need to establish clinical trial acceptance criteria involving AI. For example, pharmacists understand what the Food and Drug Administration’s clinical trial phases I, II, and III entail, she said. But “what does that mean for AI?”

Considering big questions on AI use in pharmacy

• Sikora plans to raise some of the overarching issues that must be sorted out before AI use is integrated into pharmacy. For example, what is the best way to capture medication-related data? And can that data truly improve prediction modeling?